Would it make sense to mate a Duet rcvr w external I2S DAC?

[ar-t]

So............you wanna try those 22 uF ones. Remember, their capacitance goes down as you apply voltage. It may not actually be 22 uF at 5 V DC.

Personally, I would not use a Z5U anywhere. X7R would be preferred.

One thing that you might want to try, which is impractical as all get-out in manufacturing, is to find a radial film cap that is just the right size to solder from Vcc to Vss. Yeah, you need a large junk box of parts to pull that one out. There will always be some series inductance, either in the cap and/or the traces. You are never going to get rid of all of it. But, the right size cap might not have any more inductance than you would have by using a SM cap, and then long traces to Vcc or GND.

One more thing that I should mention: ESR.

Everyone thinks that you need to have the lowest possible ESR. Well, maybe for a SMPS, yeah! That big FPGA, yeah. But on low current draw stuff?? Maybe not.

Long story short, a cap with some series R (or high ESR) is less likely to cause huge resonance spikes in impedance, if paralleled with other caps. Yes, the effective bypassing will be lower (especially at lower frequencies), but will mitigate the deleterious effects of resonances at the higher frequencies.

Another place to consider how high ESR is better is on the typical LM317-style regulator. Low ESR will resonate with output impedance of the regulator to cause a severe bump in the impedance. And also cause noise peaking. Bad for audio. Which is where you often see this type of mistake.

And not always by DIYers. Some manufacturers have never figured that out. I'll never understand how they got as far as they did. (Won't say who, but obviously no one here.)

Pat

[Pete Fowler]

Howdy Pat - glad to have you back. I've been reading like a fool and I've gathered a fair amount of information on bypass caps, digital PS design, etc. Unfortunately, info is all I've got - no wisdom or experience to make good use of all this clutter in my brain! ;-)

I could use a sprinkling of wisdom here...

One thing that you might want to try, which is impractical as all get-out in manufacturing, is to find a radial film cap that is just the right size to solder from Vcc to Vss. Yeah, you need a large junk box of parts to pull that one out. There will always be some series inductance, either in the cap and/or the traces. You are never going to get rid of all of it. But, the right size cap might not have any more inductance than you would have by using a SM cap, and then long traces to Vcc or GND.

Focusing on chip VDD pin bypass caps for a moment. Is it correct that these caps serve two purposes - 1) to bypass PS noise (and chip self-generated noise) to ground and, 2) to act as small reservoir caps to meet current demand peaks from the chip? That is my understanding at this point. I can see where the small cap/big cap combo thinking came from, and also why a low ESR large-value cap would be ideal.

(From some of your comments I'm guessing there may also be a frequency issue where the cap has to be large enough to bypass low frequency PS noise. Is this correct?)

Using the Wolfson DAC in the Duet as an example - the pairs of bypass caps on the VDD and VDA pins are filling both functions - what about the Vref pin? It has the same bypass scheme as the power pins but I would not expect any current draw from that pin. It has a 10uf lytic on it. Whassup there - shunting audio frequency noise?

Similarly, I was perusing the data sheet for the TI PCM2707 USB to I2S/SPDIF converter chip (with an onboard DAC no less) and it calls for 3 power pins to be bypassed by 100uf 'lytics! Is it safe to assume that chip makes some heavy current demands on the PS rails, hence the large value caps?

Finally, you mention using radial film caps as bypasses if they can be soldered effectively - what is it about a film cap that makes it better than a large value ceramic? Lower ESR? Larger values of capacitance? These aren't in the audio path, so?

One more thing that I should mention: ESR.

Everyone thinks that you need to have the lowest possible ESR. Well, maybe for a SMPS, yeah! That big FPGA, yeah. But on low current draw stuff?? Maybe not.

Long story short, a cap with some series R (or high ESR) is less likely to cause huge resonance spikes in impedance, if paralleled with other caps. Yes, the effective bypassing will be lower (especially at lower frequencies), but will mitigate the deleterious effects of resonances at the higher frequencies.

Another place to consider how high ESR is better is on the typical LM317-style regulator. Low ESR will resonate with output impedance of the regulator to cause a severe bump in the impedance. And also cause noise peaking. Bad for audio. Which is where you often see this type of mistake.

And not always by DIYers. Some manufacturers have never figured that out. I'll never understand how they got as far as they did. (Won't say who, but obviously no one here.)

Pat

Yes, I read a couple papers that talked about the value of a little resistance to keep things from oscillating. I use a similar trick by placing a resistor between the secondary of a power xfmr and the rectifier diodes - damps out ringing in the xfmr secondary.

So............you wanna try those 22 uF ones. Remember, their capacitance goes down as you apply voltage. It may not actually be 22 uF at 5 V DC.

Personally, I would not use a Z5U anywhere. X7R would be preferred.

Now you tell me! :-) I had decided on X7R caps but since you mentioned "high K" caps it seemed that a Z5U might be best for ESR reasons. Luckily I decided I could not afford the 22uf versions. Live and learn, ya?

Thanks again,

Pete

[ar-t]

Not sure I will answer all your questions in one go, so ask again if need be.

Big cap and little cap combo reasoning:

The physically larger cap will have more inductance than the smaller one. If you were to place two caps of different capacitances but the same size in parallel, you would not gain any advantage. The combination would have roughly the same inductance, so at the higher frequencies there is no advantage. But, if one is of a smaller size, then it raises the frequency at which the combination becomes inductive. (This may need some diagrams to explain better.)

Ok, makes sense for HF bypass. (We will forget about the resonance problems.) But, what about that noise bypass pin?

Good question. Let's think about what that pin does.

Its job is to set a reference voltage at Vcc/2. And Vcc is going to have some HF noise on it. So, there is going to be a certain amount that will get through to that pin. The 10 uF cap won't be as effective at those higher frequencies.

OK, what kind of cap to use, and why. Remember what I said about data sheets? Depends who writes them...........

This is important to understanding the "reservoir vs bypass noise" theory. At least when it comes to data sheets.

"Back in the old days, when data sheets were.........." blah, blah, etc., they would explicitly point out "This cap is recommended if the power supply is more than x inches away". It can probably be better understood if you view things from a system-wide approach, where you only have one regulator, that powers everything; or a more localised approach, where you are likely to have a separate regulator for each IC.

At one time, the guy wrote the data sheet would take all of that into account. But over time, that task was passed on to someone lower down in the chain. They viewed it as a menial task, and just did a "cut and paste" from an existing data sheet of a similar product. The pertinent info was glossed over and neglected.

More importantly, it explains the big difference in recommended implementations. Different companies have different approaches to the design process.

Which one is right?

Neither or both.

"HUH??????????? What exactly does that mean?"

There is no one singularly perfect solution to any design problem. All will have strengths and weaknesses. Engineering is always a series of compromises. You have to decide which ones that you can accept, and which ones that you can not. You pick your route, and set off to fill in the blanks. And while we may have major differences in our approaches (say one big regulator versus one at each IC), both can be made to work. They may measure differently, in which case one would hope that they sound different. They may, but different does not mean "better", just different.

Or worse..........they may measure different, yet sound about the same! (Horrors!) It can happen.

Film caps........are they better? Some think so, others do not. I'm merely encouraging you to try something different. Anything that you do will further your learning experience. Either it works, and you are encouraged. Or it doesn't, and you learned what not to try again. Of course, you may not get the same results in different situations.

I hope that does not sound like a cop-out. A lot of us learned by just those methods. We were taught little (if anything), but it was all "by the book". Unsatisfied, we wandered outside of the same confines of the sterile industrial world we were cultivated in. We came up with our own individual approaches to solving design problems, and usually invent a dogma to support it.

At one time, say 25 years ago, I worried about why caps or resistors sounded different. I am too old and too jaded to care these days. I may have ideas why things work, when implemented my way, but it could just be pure dumb luck that it works. But most importantly, I am open to learning and trying new things. Even old things again. I do use some ceramic caps, and in places where I can make a film cap work (if I think it will really work, and not just placate the insecurities of audiophiles), I use them instead.

The important thing to learn is that you have to be open to new approaches. Be just as skeptical that any suggestion will work as that it won't. Don't be afraid to expand your base. Yes, it can be a pain to buy 10 (or 100) of every part that you want to try. (That can be a whole tome for another day!)

25 years from now, we could exchange thoughts on how we both have come full-circle on our design approach. Except the odds of me being here then may not be that good. In which case, it will have to be a rhetorical debate. But then you will know why I always say "I hate this job."

Pat

[Pete Fowler]

Pat - thanks for the background info - it really helps. I can handle "it depends" as an answer. ;-)

Now to see if I understood you. Since many design tradeoffs are implementation dependent I'll use the Duet Wolfson implementation as an example:

1. Given that there is a dedicated 5V supply located approx 1" away from the DAC chip, the odds of needing 10uf at the VDD and VDA pins are small. Thus a single, lower value cap is probably better since the goal 'should be' shunting high-freq power rail and/or self-generated noise?

1a. This may be altered somewhat by the fact that the Wolfson analog output ckt is designed to drive the outputs directly, without a buffer or step up device. The analog side of the chip may need a large cap to supply extra current for those large tympani drum whacks just like a power amp supply needs juice for transients? This is a unique situation to the Wolfson.

In any case I will verify the VDD is fed by the 5V supply. If not then a large cap might be called for on that pin, too since the 3.3V supply is on the opposite side of the PCB.

2. As you mentioned, the spec sheet may be overstating the need for a large cap on the Vref pin since there is near zero power consumption at that pin. It would make sense to use a smaller cap with excellent high freq response in this case, yes?

3. By coincidence I have several Wima film caps in my junk box, but I think they're in the 150nf to 150pf range. Too small to try on the Vref bypass? Frankly, I don't have a clue how to calculate a reasonable value for a noise-only bypass cap on an IC. Audio circuits, yes. Digital no.

4. In the case where significant reservoir capacity is needed due to PS design or chip power consumption I believe a lytic is still the answer, yes? Do you have an opinion on the very small boutique lytic caps like Elna Cerafine/Silmic or Rubycon/Blackgates? They have much of the character of film caps but with higher capacitance in a small package...

As always, your insight is much appreciated!

Pete

[ar-t]

1.) There is probably a sufficient cap on the output of any regulator that you are likely to use.

1a.) And just how much current does it have to supply? (Not much, unless you are driving a 600 ohm load.)

2.) What if you have a lot of 60 Hz ripple? Not likely in the stock Duet, but think about it. And just how much noise does the stock regulator generate? Hmmm.......

How low does the noise on the Vref need to be? Think about not only level, but frequency distribution.

3.) Wima caps: I use some small ones called FKP. Yeah, low values. Also MKP2 and MKP10. Much larger, 0.1 uF size. Mouser sells some.

4.) Never tried any of the boutique 'lytics. Vishay buying up all the conventional suppliers is enough of a nightmare to deal with. BGs come in a wide array of styles, with varying ESR vs frequency characteristics. I know some guys who bought a lifetime supply before they stopped making them the first time. You would need to characterise them in order to have an idea which ones might be good in what application. But as the only place that you can find any is from "MP", and he may want $$$$$ for them. Their reputation is that they take forever to "break-in".

I have worked with designers who feel that Cerafine are horrid. Their opinion, not mine. No idea why they feel that way.

I have used solid polymer 'lytics in some applications. They may be a good choice for a digital supply, where high SRF and constant ESR would be handy. But they can be troublesome in some places. Remember my warnings on low ESR.

Pat

[Pete Fowler]

Hey Pat,

Many thanks - sorry to be slow picking this stuff up. As you say, system issues are rarely simple.

1a.) And just how much current does it have to supply? (Not much, unless you are driving a 600 ohm load.)

The spec for the Wolfson states typical analog section current draw is 9mA. Assuming some internal consumption I would guess the output driver circuit pulls 6-7mA. I suspect there won't be a problem sourcing current even with the stock 5V supply (and thanks for nudging me to think it through). ;-)

2.) What if you have a lot of 60 Hz ripple? Not likely in the stock Duet, but think about it.

Thinking out loud here - I've converted to a linear PS for my Duet. (It made a significant improvement despite claims on the forum that it wouldn't matter). It's well filtered and regulated, but there is now a chance for low freq noise coming in on the 9V supply that probably wasn't there with the original SMPS. However, that ripple is further reduced by the 5V reg (assuming good design practices) so I'm not sure PS low freq noise is an issue (at least for this implementation). In other instances - absolutely. Yes? No? Clueless?

And just how much noise does the stock regulator generate?
How low does the noise on the Vref need to be? Think about not only level, but frequency distribution.

Thinking out loud again - the answer(s) require an understanding of how the DAC works AND where the noise comes from. Looking at the Wolfson block diagram, the Vref pin is used only by the DACs - if I understand this correctly a stable Vref helps insure a stable trigger voltage on transitions (ie, if the chip triggers at +4.35V, you don't want the Vref varying so the DAC triggers at 4.3 on one bit and 4.4 on another - that is jitter as I understand it at the chip level). My guess is a very clean/stable Vref is a good thing. (Yes/no/clueless?)

Also, internally there are bit transitions at some high rate (44.1kHz?) and the resulting analog signal varying between 20-20kHz. Do these transitions generate noise that is fed back to the Vref pin? There is also the junk that comes in on the AVDD rail to deal with. The question is - what source(s) of noise is that bypass cap meant to deal with?

If one only has to deal with is PS rail noise then a clean supply and a small bypass cap should be sufficient (to shunt high freq noise). If we have to shunt internally generated 20Hz noise too, then a much larger cap is called for. At least, that's my noobish take on it.

My next step should be to hook up a 'scope and see what's actually on the Vref pin and bypass accordingly, ya? ;-) Unfortunately, I suspect my old Tek 453 won't give me the view into the problem that I would need to answer the question.

One more guess is even though noise on the Vref pin is audible it will be a subtle effect that may be near impossible to A/B detect on a macro level. I bet it's more noticable over time via listening fatigue or by a more 'digital character' to the music.

Sounds like I'm either going to have to pop for a better scope or play "swap the caps" for a month or two in hopes of stumbling on the right answer. DIY for analog is much easier...;-)

Remember my warnings on low ESR.

Got it! (I hope) Low ESR on chip pin bypass caps - usually a good thing. On most other PS bypass caps (including the outputs of reg chips) a little internal resistance is a good thing. I was reading the sheet on the LT1086 low-dropout vreg and they did a good job of spec-ing the types of caps needed and when to use them (ie, more than 2" between the PS and the vreg chip). No more sprinkling caps like holy water... 8-)

Thanks for your patience!

Pete

[ar-t]

"Thinking out loud here - I've converted to a linear PS for my Duet. (It made a significant improvement despite claims on the forum that it wouldn't matter)."

Who told you that?

"It's well filtered and regulated, but there is now a chance for low freq noise coming in on the 9V supply that probably wasn't there with the original SMPS. However, that ripple is further reduced by the 5V reg (assuming good design practices) so I'm not sure PS low freq noise is an issue (at least for this implementation)."

This won't exactly answer your question, but as usual will be more food for thought.

We have a little experiment going on over on some other forum. Some guys are listening to some Duets that I have butchered. (Sorry, Sean.............) Anyway, part of the experiment involves an outboard linear supply, and another one internal for the clock and SPDIF stuff.

We have tried several different supplies, and even though all have low output impedance (that is flat over the audio band), quiet and fairly close in design, everyone says that there are subtle differences. Why do they sound different? I dunno. I know what I did to them, and what effect it might have, but I am somewhat surprised they can tell a difference.

So, what does any of this have to do with your question?

The main difference is how we control the peak charging currents in the filter caps. No, I won't say how. (Too many "competitors" read this looking for free design help.) The point is that the little things that you might discard as being negligible can turn out to be something that is important, yet in a subtle way.

Which brings us back to ripple noise getting through to the Vref pin. You may think that it is not important, and it may well not be important. But don't discount it until you try.

I don't have a clue how that DAC chip works internally, but............

Yes, switching noise on the digital supply could spill over onto the analog side. Especially since the Duet only has one +5 V supply feeding both.

So..........more food for thought..........

Suppose, just suppose, that you put a 10R resistor in series with each supply pin. Obviously, you would want to stick a 'lytic after it to maintain DC stability. But, you now have a filter that helps keep the noise from one getting back to the other. (No, I did not try this. I am trying to keep the lid on! But no reason why it shouldn't work.)

(For the rest of you following along at home, the Duet has ferrite beads on each supply pin. Keeps RF from each other. Also prevents resonance problems that could be caused by a bunch of caps in parallel.)

OK, limiting factor on the resistor size is how low you can let the supply voltage drop. Read the data sheet very carefully before you attempt this. OK? The chip may not like one supply being a tad lower than the other. Of course, if you know the exact current draw for each, Ohm's Law can be your best friend. (Translation: 10 ohms was just a number I suggested. Both don't have to be the same.)

As for your 'scope............


Yeah, sorry, won't be a lot of help. Sure, you will see if you have switching glitches and the like, but won't help with spectral content much. If you have some small coax, you can make a cable that solders directly to the supply pin, and run ti directly to your 'scope. That 10X probe will rob you of resolution that you will probably need.

Back to ESR.........

Yes, low is good right at the pin, but..........

Somewhere up the line is a regulator. It may not be happy with it. Ah, but remember that idea about sticking a resistor in series with each supply pin? Problem solved.

Pat

[Pete Fowler]

Howdy Pat,

Again, thanks for the perspective and food for thought!

"Thinking out loud here - I've converted to a linear PS for my Duet. (It made a significant improvement despite claims on the forum that it wouldn't matter)."

Who told you that?

No one told me directly - it pops up in a couple places in older threads. The internal SMPS's are claimed to swamp any benefit of an external linear supply (or so 'tis said). Not true, fortunately.

Which brings us back to ripple noise getting through to the Vref pin. You may think that it is not important, and it may well not be important. But don't discount it until you try.

Sorry for not being clear - I absolutely think its important, just not sure how to evaluate the impact other than plain old listening (unless I win the lotto - then I might get a better 'scope). ;-)

Suppose, just suppose, that you put a 10R resistor in series with each supply pin. Obviously, you would want to stick a 'lytic after it to maintain DC stability. But, you now have a filter that helps keep the noise from one getting back to the other.

(For the rest of you following along at home, the Duet has ferrite beads on each supply pin. Keeps RF from each other. Also prevents resonance problems that could be caused by a bunch of caps in parallel.)

OK, limiting factor on the resistor size is how low you can let the supply voltage drop. Read the data sheet very carefully before you attempt this. OK?

Pat

And if I understand this correctly, the cap following the series resistor can also be a very low ESR type, and should be slightly larger than otherwise necessary (for bypassing) in order to keep the voltage sag reasonable?

Resistors in series with the pins and caps, too. Have I mentioned "hands of a surgeon" lately? Oof!

Well I have my work cut out for me. Time to order some parts and see what happens. I'm going to try the mods one at a time and listen for changes (not necessarily improvements). This is turning into a great chance to experiment (hopefully without frying the Duet in the process).

If I think of more questions may I ping you again? (And many thanks for the experience you've already shared!)

Best,

Pete

[ar-t]

"And if I understand this correctly, the cap following the series resistor can also be a very low ESR type, and should be slightly larger than otherwise necessary (for bypassing) in order to keep the voltage sag reasonable?"

Yes! This is one possible approach. Doesn't mean that it is "more correct" than some other approach, merely something to think about. I know some really expensive (and highly reviewed) gear that operates this way. And there probably some really awful stuff done this way.

Back to the Vref bit.......

Didn't say that you stated that it was not important. Merely trying to point out things that seem minor often turn out to be more important. Yes, these sort of tweaks may only make subtle changes, but when you add them all up, then you get something that could be special.

One last thing to fret over.

A lot of hobbyists get frustrated when folks like me don't give them schematics on how to do this stuff. Two problems with that:

1.) You don't learn diddly, and therefore will never really progress any. You can't ride someone else's coat tails forever. They don't like it when folks like me throw that one in their face.

2.) All the schematics in the world will not assure stellar results. How it is implemented is also part of the equation. There is no way to explain that in so many words on a forum.

So, I can give you some off-the-wall ideas, which get your brain working. Eventually, it will all sink in, you will try it (and other stuff as well), and then you will be able to haul the load all by yourself.

Doing good so far.

Pat

[Pete Fowler]

Howdy Pat,

Well, I might have the hands of a surgeon after all - didn't fry anything (although I did lift one pad. Dang!).

I've done some work around the DAC chip and thought I should check in to see if what I'm hearing makes sense (and since my 'scope isn't so good, hearing is all I have)...

Basically I decoupled the DVDD and AVDD pins with 15ohm 1/8W metal film resistors (I would have used 1/16W if I could find them - smaller is better around those chips). Then I bypassed the power pins directly to their respective ground pins with 1.0uf X7R ceramics. I thought 1uf would be sufficient to deal with the voltage sag across the resistors. Used very low ESR Murata's.

Finally I bypassed the Vref pin to the analog ground with...a 10uf/50V Z5U ceramic. A risk, but since this is for educational purposes I figured 'why not?' The part is spec'ed at 50V and the pin only has 2.5V on it so I'm not so worried about losing capacitance with voltage. As always, I'm guessing. ;-)

And that's it. I tried to do things in stages, ie, the AVDD pin mods first, etc. But with the tight fit I ended up doing the DVDD mods and the Vref mod at the same time.

And the result?

With the AVDD pin mod (only) the highs improved a bit but the big change was the soundstage opened up. Big and 3D. Not subtle either. Lots of depth and the field extended out past the speakers just like my vinyl rig will on occasion. Neat.

Then I added the DVDD and Vref mods and the big change there was improved tone. A better description might be - there was more "there" there. Piano sounds fuller and tonally richer. And the bass improved in quantity (but not quality, oddly enough). Still, quite impressive given I haven't messed with the power supply, clocks or anything radical like that.

So if I were to describe the Duet's sound with a linear external PS and the decoupling/bypassing mods I would say it's presenting a lot more information (tone, soundstage, bass) than it was, but everything is still not well focused.

Like seeing a very rich and detailed painting through dirty lenses. I know the music's in there but it's furry. The highs are crunchy and the bass is fat too. (This is all relative by the way - it sounds pretty good as-is).

True confession time - I also changed the output coupling caps as part of the process. No real change except maybe the highs got a bit smoother. But they didn't change the fundamental sound of the unit much. Very interesting...

Other than adding a cleaner 5V supply to the DAC chip I think I've milked that part of the circuit for all it's worth. I'd like to try swapping Vref bypass caps to see what happens but it's a tight fit and I'm afraid I'd do damage in the process.

Maybe a second duet...;-)

Next I'm going to noodle on the 11.xxxMhz crystal and inverter chip - noisy combo, that. And perhaps a cleaner 5V supply. Thinking is (mostly) free and I don't lift pads.

Thanks again for your guidance - this has been really eye-opening.

Pete